We will survey recent AI regulatory and policy activities from several perspectives, to understand their potential for helping the development of artificial moral agency, or Phronesis.
Summary What sort of ethics framework would be needed as a precursor to pursuing the goal of machine consciousness? Are there insights to be gleaned from current policy developments in the field of ethics in the context of current and near-future AI systems? This talk will survey recent regulatory activities in this arena from several perspectives and examine how these might – or might not – help us to explore the development of artificial moral agency as a potential contribution to the mechanistic underpinnings of future machine consciousness.
There are several salient elements of relevance here. One is the
need to ensure that ethical considerations are incorporated in the initial
specification, engineering design, and development of AI systems.
Some of the key concerns here are the need to recognize inherent biases
and problematic strategies, such as those related to the collection and/or
selection of training data for machine learning systems, for example
Microsoft’s ill-considered Tay system.[
A second element is the application of ethical principles during the
prototyping/testing of AI systems. There is a significant concern that,
as autonomous systems get more powerful and adept at human tasks,
the corresponding ethical considerations scale as exponentially large as
the applications. The problem is that, when major AI development
organizations put ethics on the backburner, major risks can spiral up and
out of control. Of particular interest here is the purposeful incorporation
of human participants – who themselves may or may not be aware of
their involvement – in the testing and validation of AI systems.[
Separately from these elements, which are applicable to most or all
human interaction and decision support systems, we also need to
concern ourselves with the challenges involved with designing a concept of
2
ethics awareness into the AI products themselves. The vision of
building artificial moral agents, which operationalize moral practical
wisdom or Aristotelian Phronesis, falls under this category.[
One way forward is to develop a corpus of case studies, which can
be used to guide system designers and researchers to envision
techniques for architecting such agents. For a very simplistic example, we
can look at the European Union’s introduction of the General Data
Protection Regulation (GDPR), which directs how an individual’s personal
data should be handled and processed.[
The initial benefit of such a body of case law will enable individuals to challenge the data practices of organizations, and in turn allow organizations to take data protection authorities to task over their enforcement actions. However, in the long term, this corpus also provides a valuable resource for characterizing tradeoffs between individual rights vs. common goods, implicit vs. explicit consent, and other issues in the areas like privacy, anonymity, security, etc. In other words, a body of structured examples of how ethical dilemmas and tradeoffs are actually resolved in practice. In this way, such a corpus could eventually become a key resource for developers of artificial phronesis.